[R-sig-eco] Error in lm AND questions on betadisper()

[Martin Weiser]

On 08. 02. 22 12:24, Ciraolo, Alessia Caterina wrote:

> 
> 2) I need to run a PERMANOVA analysis with interaction term . However
> before that, I was testing the assumptions with betadisper but I have
> trouble in making the variable "group" in betadisper. Usually, it usually
> consists in a factor or an object with a single level, is it right? But how
> does it work with the interaction term, please?
> Below you can find my code:
> mod3 <- betadisper(distance, group)
> mod3
> anova(mod3)
> 
> permutest(mod3, pairwise = TRUE, permutations = 999)
> 
> perm <- adonis(distance ~ Habitat *Treat, data=BF,
>                 permutations= 9999, method="euclidean")
> perm
> 

Dear Alessia,

I cannot address the first issue, but with the second I suggest You make 
groups for the Habitat and Treatment combinations (if they are factors, 
of course). So, with habitats A-C and treatments tr1-tr3 you get a new 
factor for grouping, with nine levels: Atr1, Atr2, Atr3, Btr1,...,Ctr3.

And with this, I am much less sure: if one of the pair Habitat/Treatment 
is continuous var. I am not sure - but I would first "extract/offset" 
the signal of the continuous variable, for example by running RDA with 
the continuous variable as the constraint (predictor), using resulting 
scores on the non-constrained axes as the new data to be grouped 
according to the levels of the factor variable. Or, I would detrend each 
variable of the response data separately (running lm with the continuous 
variable (Treat or Hab) as the predictor, and using residuals instead of 
original values). I am afraid that this is not great, because if the 
continuous and the factor variable share some explained variability, you 
force it to "belong" to the continuous predictor.

HTH,
Martin